Estimating missing data in historic series of global radiation through neural network algorithms

Abstract

In data processing time series of meteorological data problems, you are incomplete in some time intervals; it addresses the issue commonly using the autoregressive integrated moving average (ARIMA) or the method by regression analysis (interpolation), both with certain limitations under particular conditions. This paper presents the results of an investigation aimed at solving the problem using neural networks reported. The analysis of a time series of global radiation obtained at the Francisco de Paula Santander University (UFPS) is presented, with basis in the recorded data by the weather station attached to the Department of Fluids and Thermals. Having a series of ten-year study for 125,658 records of temperature, radiation and energy with a percentage of 9.98 missing data, which were duly cleared and completed by a neural network using algorithms backpropagation in the mathematical software MATLAB